A Comparative Study of Tensor Networks and Traditional Neural Networks

Abstract

Tensor networks with various structures have been proposed as alternatives to traditional neural networks. Tensor networks represent low-rank factorizations of traditional neural network kernels and can be evaluated using fewer arithmetic operations. However, higher performance is not guaranteed since the achievable fraction of machine peak for evaluation of tensor networks with many small tensors can be considerably lower than that attained in evaluating traditional neural network kernels. This project will preform a comparative evaluation of currently achievable performance on multicore CPUs as well as GPUs, using traditional kernels versus tensor networks. a number of popular machine learning frameworks like TensorFlow and PyTorch will be evaluated, as well as stand-alone code generators and libraries for tensor computations.

Document Details

Document Type
DoD Grant Award
Publication Date
Sep 04, 2019
Source ID
W911NF1910491

Entities

People

  • Ponnuswamy Sadayappan

Organizations

  • Army Contracting Command
  • National Security Agency
  • University of Utah

Tags

Fields of Study

  • Computer science

Readers

  • Neural Network Machine Learning.
  • Parallel and Distributed Computing.

Technology Areas

  • AI & ML
  • AI & ML - Machine Learning Algorithms
  • AI & ML - Neural Networks